Accelerated brain aging associated with executive deficits in adolescence and early adulthood among individuals with schizophrenia and genetic high risk

Abstract

Introduction

Abnormal brain development and cognitive function are present in adolescence and early adulthood of schizophrenia (SZ). However, it is still unclear whether these developmental abnormalities and cognitive deficits occur in adolescence and early adulthood in individuals at genetic high risk (GHR). This study aims to examine brain age gap differences during adolescence and early adulthood in individuals with SZ, GHR, and healthy control (HC), as well as their associations with cognitive function.

Methods

This study recruited a total of 240 subjects aged between 12 and 25, comprising 83 SZ, 48 GHR, and 109 HC. We calculated all participants’ brain age gap using their structural MRI. The subtests of Wisconsin Card Sorting Test were utilized to evaluate executive function.

Results

Compared to HC group, both GHR group (P = .021) and SZ (P < .001) group showed greater brain age gap (brain age > chronological age). Notably, within GHR group, greater brain age gap was associated with higher non-perseverative errors in executive function (r=.39, P_FDR=0.035), whereas in SZ group, it was correlated with higher perseverative errors (r=.35, P_FDR=0.005).

Conclusions

This study suggests that changes in brain age during adolescence and early adulthood in SZ and GHR individuals may reflect disease’s genetic susceptibility. A greater brain age gap may be associated with the onset and progression of executive deficits in SZ.

Introduction

Schizophrenia (SZ) is a highly heritable neurodevelopmental disorder, and its onset may be linked to abnormal brain changes resulting from excessive synaptic pruning [1, 2]. Such developments predominantly take place from adolescence to early adulthood, a critical period for brain maturation that also coincides with the highest occurrence of the disease [3, 4]. Additionally, unaffected family relatives of patients with SZ have been found to face a markedly higher risk of the disorder [5] and show subtle changes in neurodevelopment [6]. Research on individuals with SZ and genetic high risk (GHR) is essential for understanding neurodevelopmental processes in SZ.

Recent advances have introduced the concept of brain age to better capture disease effects and abnormal brain development [7–9], that is, using machine learning models to predict age from magnetic resonance imaging (MRI), and the differences between brain age and chronological age is called brain age gap estimation (BrainAGE). Structural MRI studies, both cross-sectional and longitudinal, have revealed that individuals with SZ often show a significantly increased BrainAGE [10, 11], which is especially pronounced in chronic cases within five years of onset [12]. In the early course of SZ, a greater BrainAGE is correlated with longer disease duration. Individuals with clinical high-risk for SZ who later develop psychosis exhibit a higher BrainAGE than those who did not convert [13]. Another research showed that brain age is a significant predictor of psychosis conversion [14]. This underscores the critical importance of assessing brain age during the disease’s early phases, even before symptom onset. While studies have examined brain age in GHR [15, 16], it remains unclear whether such individuals exhibit brain age differences during adolescence and early adulthood brain development.

Cognitive deficits are a core feature of SZ [17], affecting various aspects of neuropsychological function, particularly in executive functions [18]. As a central component of higher-order cognitive control, executive dysfunction may further impair overall cognitive performance [19] and significantly influence patients’ social functional prognosis [20, 21]. Executive dysfunction may prominently involve impairments in cognitive flexibility, the ability to adaptively shift mental sets in response to changing contingencies, which can be assessed by Wisconsin Card Sorting Test (WCST). Crucially, cognitive flexibility is uniquely vulnerable to processing speed mediation, wherein slowed information processing fully underlies set-shifting deficits [19], which are observable not only in SZ but also as mild alterations in GHR [22–24]. These cognitive deficits emerging in adolescence and early adulthood are particularly noteworthy, as they may signal an increased risk for psychosis [25]. This likely reflects a deviation from the normal trajectory of neurodevelopment, which can be assessed by brain age differences. However, research on whether executive deficits are related to brain age differences in adolescence and early adulthood is limited.

In summary, in our study, we assessed executive function and brain maturation during adolescence and early adulthood across three groups: healthy control (HC), patients with SZ, and individuals at GHR. We hypothesized that: (1) both SZ and GHR groups would display larger brain age gaps compared to HC; (2) greater brain age gaps in SZ and GHR groups would be scaled with reduced executive function.

Methods

Participants

This study enrolled 240 participants aged 12–25 years, including 83 SZ patients, 48 individuals with GHR, and 109 HC. Patients diagnosed with SZ were recruited from two clinical facilities: the First Affiliated Hospital of China Medical University and the Shenyang Mental Health Centre. The GHR group was comprised solely of first-degree relatives of SZ patients, enrolled from the same institutions, while HC group was drawn from the local community through targeted advertising efforts. Trained psychiatrists conducted all evaluations following the protocol of the Structured Clinical Interview for DSM-IV-TR Axis I Disorders (SCID-I). Participants in the SZ group met the DSM-IV-TR diagnostic standards, whereas individuals in GHR group had no recorded history of any Axis I disorders, and HC reported no such disorders either personally or within their families. Additionally, GHR group was further assessed with the Structured Interview for Prodromal Syndromes (SIPS) to ensure the absence of prodromal psychotic symptoms.

All participants had to satisfy stringent exclusion criteria, which encompassed: (1) a history of substance abuse or dependency, including alcohol; (2) any serious medical conditions; (3) neurological disorders; (4) any previous head injury that resulted in a loss of consciousness for at least five minutes; (5) any contraindications to MRI scans; and (6) instances where the MRI data were not of acceptable quality.

Psychiatric symptom severity was assessed using standardized instruments: Brief psychiatric rating scale (BPRS) [26], 17-item Hamilton Rating Scale for Depression (HAMD-17) [27]and Hamilton Rating Scale for Anxiety (HAMA) [28]. To evaluate executive function, the Wisconsin Card Sorting Test (WCST) was employed, gathering data on five variables including the number of correct responses (CR), categories completed (CC), total errors (TE), perseverative errors (PE), and non-perseverative errors (NPE).

All procedures in this study adhered to ethical standards established by relevant national and institutional committees responsible for human research, as well as to the guidelines of the Helsinki Declaration of 1975 (revised in 2008). The study protocol was approved by the Medical Research Ethics Committee of the First Hospital of China Medical University (ethical approval number: [2012] 25–1), and all participants provided written informed consent.

Structural MRI acquisition and processing

T1-weighted images were obtained from all subjects using a 3D Fast Spoiled Gradient Echo (3D-FSPGR) sequence on a Signa HDx 3.0 Tesla MRI scanner. The protocol was configured with a repetition time (TR) of 7.1 ms, an echo time (TE) of 3.2 ms, and a flip angle of 13°. Moreover, the field of view was established at 240 mm × 240 mm, capturing 176 slices at a thickness of 1 mm each. This produced isotropic voxels measuring 1.0 × 1.0 × 1.0 mm³ arranged on a 240 × 240 matrix, with the entire scanning process lasting 8 min and 22 s.

Image processing was conducted with FreeSurfer 7.4.1 [29], which automatically segmented cortical regions using the Desikan-Killiany atlas [30] and identified subcortical structures based on FreeSurfer’s atlas [31]. Three morphological metrics were derived: cortical thickness and surface area (measured across 68 regions, with 34 per hemisphere), and subcortical volumes (across 14 regions, with 7 per hemisphere).

Brain age gap estimation

The Centile BrainAGE algorithm was applied to assess the brain age gap, which refers to the gap between an individual’s estimated brain age and their actual chronological age. This machine learning model builds on earlier approaches for predicting BrainAGE in young populations [32], namely, a nonlinear support vector regression with radial basis function kernels [33], and was developed and validated using data from over 40,000 participants between the ages of 5 and 90, as part of the Enhancing Neuro Imaging Genetics by Meta-Analysis (ENIGMA) consortium [34]. Notably, the tool’s applicable range (5 to < 40 years) accommodates the younger age group present in our study sample (ages 12–25).

Our data, as an independent test set, were provided to the online Centile BrainAGE tool, including sex, age, 68 unilateral cortical thickness and 68 unilateral cortical surface area metrics from the Desikan-Killiany atlas, as well as 14 unilateral subcortical volume measures derived from the FreeSurfer atlas. Previous studies [35, 36] have noted that brain age tends to be overestimated in younger individuals and underestimated in older ones, with optimal accuracy observed for ages close to the average of the training dataset. To address this systematic age bias, the Centile BrainAGE model applies a correction technique developed by Beheshti et al. [37]. Importantly, the CentileBrain BrainAGE models will optimize accuracy by using a combination of linear and non-linear functions that are customized to each uploaded dataset by estimating the adjustment from the observed age-prediction relationship in that dataset, then applying it to generate an adjusted predicted age.

After the user inputs data, the online tool will provide an estimation of brain age that has been adjusted for bias. The bias-corrected brain age gap estimation (BrainAGE_corr) values are then computed by subtracting the individual’s chronological age from this bias-corrected brain age estimation. Positive BrainAGE_corr values reflect that brain age is higher than chronological age. Conversely, negative BrainAGE_corr values indicate that brain age is lower than chronological age. All analyses in this study utilize these bias-corrected BrainAGE_corr values. To assess the algorithm’s performance across our sample, we calculated the mean absolute error (MAE) between the estimated brain age and the chronological age, separately for each by group and sex.

Statistical analysis

Group differences in BrainAGE_corr values and other continuous variables (demographic, clinical, and cognitive characteristics) were tested using analysis of variance (ANOVA) with LSD post-hoc comparisons, whereas categorical variables were compared using the chi-square test. Analysis of covariance (ANCOVA) was also utilized to examine whether significant group differences in BrainAGE_corr values persisted, even after adjusting for potential demographic confounders (age, sex, education years and BMI). We examined clinical correlations in SZ patients between BrainAGE_corr values and symptom severity (HAMD-17, HAMA, BPRS), illness duration, and WCST performance using Pearson correlations. We also conducted a correlation analysis between BrainAGE_corr values and the WCST items in the GHR group. False Discovery Rate (FDR) was employed to control for multiple comparisons. (P_FDR < 0.05 significant).

Results

Demographic, clinical, and cognitive characteristics

No significant differences were observed among groups in terms of age or BMI. However, a significant disparity was noted in sex and years of education. Additionally, the SZ group demonstrated markedly higher scores on the BPRS, HAMD-17, and HAMA assessments relative to GHR and HC groups. On CC and CR metrics of the WCST, both the HC group and the GHR group performed better than the SZ group. Conversely, the SZ group scored highest on TE and PE metrics, and no significant differences were observed in these four metrics between the GHR and HC groups. Only the NPE metric was the highest in the SZ group, followed by the GHR and HC groups (Table 1).

Table 1.

Demographic, clinical and cognitive characteristics of participant groups

	HC (n = 109) Mean ± SD	GHR (n = 48) Mean ± SD	SZ (n = 83) Mean ± SD	F/χ2	P	Post hoc analysis
Demographics
Age, (range: 12–25)	19.03 ± 3.83	18.88 ± 3.83	18.11 ± 3.31	1.58	0.209	-
Sex (% male)	54.13	70.83	44.58	8.44	0.015	-
Education years	13.30 ± 3.72	11.56 ± 2.84	10.41 ± 2.57	19.82	< 0.001	HC > GHR> SZ*
BMI	21.93 ± 4.10	21.52 ± 3.92	22.83 ± 3.75	2.03	0.134	-
Clinical characteristics
Medication, yes	--	--	70	-	-	-
Duration of illness (months)	--	--	28.60 ± 27.18	-	-	-
HAMD-17, total scorea	1.61 ± 2.37	1.33 ± 2.17	6.13 ± 6.90	28.14	< 0.001*	HC = GHR< SZ*
HAMA, total scorea	1.49 ± 2.52	0.80 ± 1.20	6.12 ± 7.20	29.64	< 0.001*	HC = GHR< SZ*
BPRS, total scoreb	18.89 ± 1.75	18.57 ± 1.60	33.07 ± 13.82	74.45	< 0.001*	HC = GHR< SZ*
Wisconsin Card Sorting Test
Correct responses (CR)	33.81 ± 8.74	32.02 ± 8.10	23.04 ± 11.16	31.82	< 0.001*	HC = GHR> SZ*
Categories completed (CC)	5.11 ± 4.21	4.04 ± 1.65	2.65 ± 2.79	12.61	< 0.001*	HC = GHR> SZ*
Total errors (TE)	14.22 ± 8.67	15.98 ± 8.10	25.31 ± 10.82	35.08	< 0.001*	HC = GHR< SZ*
Perseverative errors (PE)	5.82 ± 5.70	4.83 ± 3.87	8.59 ± 5.13	10.14	< 0.001*	HC = GHR< SZ*
Non-perseverative errors (NPE)	8.61 ± 4.96	10.90 ± 5.74	16.73 ± 7.66	41.54	< 0.001*	HC < GHR< SZ*

Note: F: ANOVA; χ²: Chi-square test; SD: Standard deviation; BMI, Body Mass Index; HC, Healthy Control; SZ, Schizophrenia; GHR, Genetic High Risk; BPRS, Brief Psychiatric Rating Scale; HAMD-17, the 17-item version of the Hamilton Rating Scale for Depression; HAMA, Hamilton Rating Scale for Anxiety

*: P-values were corrected for multiple comparisons using FDR method

^aHAMD-17 total score and HAMA Total Score for 2 GHR, and 8 SZ participants were missing

^bBPRS total score for 13 HC,1 GHR, and 1 SZ participants were missing

Assessment of the Brain age algorithm

MAE is a common indicator for assessing the performance of brain age models. The MAE were 2.79 years for males and 2.72 years for females in the HC group, and 2.72 years for males and 1.98 years for females in the GHR group, and 4.00 years for males and 5.01 years for females in the SZ group.

Comparisons of Brain age gap estimation between groups

In our sample, the average BrainAGE_corr values were 0.63 ± 3.36, 1.90 ± 2.73, 4.29 ± 3.10, and for HC, GHR and SZ groups, respectively. ANOVA revealed significant differences among these groups (F [2,237] = 31.91, P < .001, eta² = 0.21, 95% CI [0.124, 0.295]; Fig. 1). Specifically, SZ group showed notably greater BrainAGE_corr values than HC (+ 3.66 years, P < .001) group and GHR (+ 2.39 years, P < .001) group, while GHR group also showed significantly greater values than HC group (+ 1.27 years, P = .021).

Fig. 1.

(a) Estimated Brain Age_corr vs chronological age for healthy control (HC), genetic high risk (GHR), and schizophrenia (SZ) groups. A dashed line with a slope of 1 is used as a reference. (b) Bias-corrected brain age gap (i.e., BrainAGE_corr) for HC, GHR, and SZ groups. The box plot illustrates the mean values along with their corresponding standard deviations. *P < .05, ***P < .001

Brain age gap differences in groups accounting for demographic covariates

When controlling for age, sex, education years, and BMI in an ANCOVA, significant group differences in BrainAGE_corr values among three groups were also observed (F [2,233] = 18.86, P < .001, eta² = 0.14). Specifically, SZ group showed greater BrainAGE_corr values than HC group (+ 3.00 years; P < .001) and GHR group (+ 1.89 years; P = .001). Moreover, a significant difference was also found between GHR group and HC group (+ 1.11 years; P = .040). Additionally, greater BrainAGE_corr values were significantly correlated with age (P = .006), but no significant associations were found with sex (P = .288), education years (P = .208), or BMI (P = .353).

Clinical and cognitive measures associations with Brain age gap

The study did not find significant links in SZ between BrainAGE_corr values and several clinical measures (HAMD-17, HAMA, BPRS, and duration of illness) (Table 2). In terms of cognition, in the SZ group, greater BrainAGE_corr values were associated with an increase in perseverative errors (r = .35, P_FDR = 0.005; Fig. 2). Additionally, we also examined cognitive measures in GHR individuals to see whether they are associated with BrainAGE_corr values. It showed that greater BrainAGE_corr values were associated with an increase in non-perseverative errors (r = .39, P_FDR = 0.035; Fig. 3).

Table 2.

Bias-adjusted Brain age gap correlations with clinical and cognitive measures in schizophrenia and genetic high risk group

	r	P FDR
Clinical measures (Schizophrenia group)
HAMD-17, total score	0.07	0.565
HAMA, total score	0.12	0.405
BPRS, total score	0.12	0.405
Duration of illness	− 0.19	0.367
Wisconsin Card Sorting Test (Schizophrenia group)
Correct responses (CR)	− 0.20	0.117
Categories completed (CC)	− 0.19	0.117
Total errors (TE)	0.19	0.117
Perseverative errors (PE)	0.35	0.005
Non-perseverative errors (NPE)	0.03	0.814
Wisconsin Card Sorting Test (Genetic high risk group)
Correct responses (CR)	− 0.31	0.055
Categories completed (CC)	− 0.29	0.055
Total errors (TE)	0.31	0.055
Perseverative errors (PE)	0.12	0.421
Non-perseverative errors (NPE)	0.39	0.035

Note: P_FDR: P-values were corrected for multiple comparisons using FDR method

Fig. 2.

Correlation between Brain age gap and executive function in schizophrenia patients. Scatter plot showed schizophrenia group’s BrainAGE_corr values were positively correlated to Perseverative Errors (r = .35, P_FDR = 0.005)

Fig. 3.

Correlation between Brain age gap and executive function in individuals with genetic high risk. Scatter plot showed genetic high risk group’s BrainAGE_corr values were positively correlated to Non-perseverative Errors (r = .39, P_FDR = 0.035)

Discussion

This research shows that accelerated brain aging occurs in SZ patients and individuals with GHR between the ages from 12 to 25 and supports that BrainAGE not only reflects genetic susceptibility but also the concept of BrainAGE as an intrinsic phenotype dependent on developmental stage. Additionally, our study demonstrates that accelerated brain aging is linked to deficits in executive function, as shown by distinct associations: a positive correlation with PE in the SZ group and with NPE in the GHR group.

Adolescence and early adulthood, a critical neurodevelopmental stage characterized by dynamic changes in grey matter density [38, 39], synaptic pruning [2, 40], coincide with the peak period of SZ onset [3]. Previous studies examining brain age in adult SZ patients and their first-degree relatives failed to detect significant brain age acceleration in individuals at GHR [15, 16]. This may stem from the broad age distribution in prior cohorts, wherein compensatory mechanisms in adulthood could mask subtle developmental lags present during adolescence [41–43], and this mechanism becomes evident in the late stages of adolescence [42]. In contrast, our study focused on a younger age range (12–25 years), revealing a modest yet significant familial tendency toward accelerated brain aging in the GHR group.

Our findings suggest that the association between whole-brain age differences and executive function exhibits disease stage specificity. In SZ patients, the association between greater brain age gap and increased PE may be attributed to the shifting ability, a key component of PE [44], is primarily mediated by a slowing of processing speed, and age modulates this mediation, which can be understood as accelerated aging [45]. In contrast, in GHR individuals, the association between greater brain age gap and increased NPE seems to suggest that subtle neurodevelopmental deviations in adolescence and early adulthood may manifest as abnormalities in the sustained attention network [46, 47], consistent with NPE being considered as a reflection of difficulties in sustained attention [48–50]. Combined with our findings, existing evidence [51] suggests that greater brain age differences could be associated with the onset and progression of executive deficits in SZ.

Based on the findings of this study, even though GHR group did not exhibit any psychiatric or emotional symptoms, they demonstrated mild deficits in cognitive executive function. This suggests the need for early screening of this population, such as through functional connectivity imaging [52, 53], and the implementation of early interventions similar to individualized patient treatment to reduce their risk of disease onset [54] or develop into cognitive impairments [24]. This is because characteristics in the cognitive domain, such as executive function deficits, typically do not improve with existing treatments [17, 55].

This study has several limitations. Notably, cross-sectional design restricts the ability to determine causal links between brain age acceleration and executive dysfunction among GHR. Second, although we controlled for key demographic covariates (age, sex, education years and BMI), unmeasured confounding factors, such as socioeconomic status, prenatal exposures, or pubertal hormone levels, may influence both brain aging and cognitive performance. Thirdly, future research should further construct and apply interpretable brain age models (i.e., regional brain age models) to reveal the spatial specificity and neural mechanisms behind accelerated brain aging. Finally, our brain age estimation relied on structural MRI metrics alone. Incorporating multimodal neuroimaging data could enhance the biological specificity of brain age models and clarify whether accelerated aging reflects global versus network-specific deviations.

Conclusion

In summary, our findings suggest that accelerated brain aging during adolescence and early adulthood may reflect genetic susceptibility. A greater brain age gap could be associated with the onset and progression of executive deficits in SZ.

Abbreviations

SZ

Schizophrenia

GHR

Genetic high risk

HC

Healthy control

BrainAGE

Brain age gap estimation

BrainAGE_corr

Bias-corrected brain age gap estimation

WCST

Wisconsin Card Sorting Test

CR

Correct responses

CC

Categories completed

TE

Total errors

PE

Perseverative errors

NPE

Non-perseverative errors

BPRS

Brief psychiatric rating scale

HAMA

Hamilton Rating Scale for Anxiety

HAMD-17

The 17-item Hamilton Rating Scale for Depression

MRI

Magnetic resonance imaging

MAE

Mean absolute error

ENIGMA

Enhancing Neuro Imaging Genetics by Meta-Analysis

FDR

False discovery rate

ANOVA

Analysis of variance

BMI

Body mass index

ANCOVA

Analysis of covariance

SIPS

Structured Interview for Prodromal Syndromes

Author contributions

All contributors have thoroughly reviewed and given their approval for the final version of the manuscript. Furthermore, each contributor fulfills the authorship criteria outlined in the Uniform Requirements for Manuscripts Submitted to Biomedical Journals, as follows: Study concept and design: CD and YT. Acquisition of data: CD, PW, LK, YZ1, TS, ZW, YZ2, QS, ZT, LW, YL, YS and YT. Analysis and interpretation of data: CD and PW. Drafting of the manuscript: CD and YT. Critical revision of the manuscript for important intellectual content: CD, PW, LK, YZ1 and YT.

Funding

This work was supported by the Joint Fund for Regional Innovative Development (U24A20700, to YT), the Science and Technology Innovation 2030-Major Project on Brain Science and Brain-like Research (2021ZD0200700, to YT), the Clinical Application and Imaging-Genetic Research of Brain Age Prediction Models for Schizophrenia(2023JH2/101300031,to YT) and Basic Scientific Research Projects of Universities of Liaoning Province (LJKMZ20221214 to Yifang Zhou).

Data availability

Data used in this study can be accessed on proper requirement from the corresponding authors.

Declarations

Ethics approval and consent to participate

This study received ethical clearance from the Medical Science Research Ethics Committee of the First Affiliated Hospital of China Medical University ([2012]25 − 1), and written informed consent was obtained from all participants. For individuals under the age of 18, written informed consent was also obtained consent from their parents or legal guardians. Clinical trial number: not applicable.

Consent for publication

Not applicable.

Competing Interests

The authors declare no competing interests.